diff --git a/binaries/raspberry_casket.bin b/binaries/raspberry_casket.bin
index 66f9908..3e17934 100644
Binary files a/binaries/raspberry_casket.bin and b/binaries/raspberry_casket.bin differ
diff --git a/src/raspberry_casket.asm b/src/raspberry_casket.asm
index c50f76f..f61258d 100755
--- a/src/raspberry_casket.asm
+++ b/src/raspberry_casket.asm
@@ -1623,8 +1623,8 @@ pre_PlayerTick:
         move.w  d5,pcd_out_len_w(a5)
 
         moveq.l #-1,d3
-.inst_set_wave_has_subloop
         move.w  d3,pcd_out_lof_w(a5)
+.inst_set_wave_has_subloop
         move.l  a1,pcd_out_ptr_l(a5)
         move.w  d1,pcd_inst_loop_offset_w(a5)
 
@@ -1930,15 +1930,19 @@ pre_PlayerTick:
 ; ----------------------------------------
 ; wave loop pos advancing and pattern wave offset command handling
 
+        moveq.l #0,d1
+        tst.b   wi_allow_9xx_b(a3)
+        sne     d1
+        and.b   pcd_wave_offset_b(a5),d1
+
         move.w  wi_subloop_len_w(a3),d3
         beq     .wave_has_no_subloop
         move.w  d3,pcd_out_len_w(a5)        ; FIXME can we move this to wave loading?
+        move.w  pcd_inst_subloop_wait_w(a5),d5
 
+        move.b  pcd_inst_ping_pong_dir_b(a5),d4
         move.w  wi_subloop_step_w(a3),d2
-        moveq.l #0,d1
-        move.b  pcd_wave_offset_b(a5),d1
-        beq.s   .wave_with_subloop_but_no_wave_offset
-        tst.b   wi_allow_9xx_b(a3)
+        tst.w   d1
         beq.s   .wave_with_subloop_but_no_wave_offset
 
         ; update loop offset from pattern
@@ -1947,7 +1951,7 @@ pre_PlayerTick:
         clr.b   pcd_wave_offset_b(a5)
 
         ; keep current direction of ping-pong unchanged
-        move.b  pcd_inst_ping_pong_dir_b(a5),d4
+        tst.b   d4
         bpl.s   .wave_move_one_step_ahead
         sub.w   d2,d1       ; go in reverse direction one step?
         bra.s   .wave_submove_cont
@@ -1957,23 +1961,29 @@ pre_PlayerTick:
 
 .wave_with_subloop_but_no_wave_offset
         move.w  pcd_inst_loop_offset_w(a5),d1
-        subq.w  #1,pcd_inst_subloop_wait_w(a5)
+        subq.w  #1,d5
         bgt.s   .wave_subloop_wait
 
         ; subloop moves!
-        move.b  pcd_inst_ping_pong_dir_b(a5),d4
 .wave_submove_cont
         ; reset subloop wait
         moveq.l #0,d5
         move.b  wi_subloop_wait_b(a3),d5        ; FIXME why is this not increased by one here?
-        move.w  d5,pcd_inst_subloop_wait_w(a5)
 
         tst.b   d4
-        bpl.s   .loop_is_moving_backwards
+        bmi.s   .loop_is_moving_forwards
+        sub.w   d2,d1                   ; decrement offset in backward direction one step
+
+        move.w  wi_loop_start_w(a3),d2
+        sub.w   d1,d2                   ; calc how many bytes we are past front
+        bmi.s   .wave_new_loop_pos_fits
+        bra.s   .wave_loop_dir_changed
+
+.loop_is_moving_forwards
         add.w   d2,d1                   ; increment offset in forward direction one step
 
         move.w  d1,d4
-        add.w   d3,d4                   ; calculate next start of loop (curr start)
+        add.w   d3,d4                   ; calculate new end position of loop sample
 
         move.w  wi_loop_end_w(a3),d2
         cmp.w   d1,d2
@@ -1983,46 +1993,28 @@ pre_PlayerTick:
         sub.w   d4,d2                   ; space left = (max(loop end, sample end) - curr start)
         bhi.s   .wave_new_loop_pos_fits ; max(loop end, sample end) > curr start?
 
-        add.w   d2,d1                   ; place offset at last possible place (loop_end - loop size)
-
-        clr.b   pcd_inst_ping_pong_dir_b(a5) ; mark going backwards
-        bra.s   .wave_loop_dir_changed
-
-.loop_is_moving_backwards
-        sub.w   d2,d1                   ; decrement offset in backward direction one step
-
-        move.w  wi_loop_start_w(a3),d4
-
-        move.w  d4,d2
-        sub.w   d1,d2                   ; calc how many bytes we are past front
-        bmi.s   .wave_new_loop_pos_fits
-
-.wave_hit_front_of_loop
-        move.w  d4,d1
-        st      pcd_inst_ping_pong_dir_b(a5)
-
 .wave_loop_dir_changed
+        add.w   d2,d1                   ; fix front of looip
+        not.b   pcd_inst_ping_pong_dir_b(a5)
+        IFEQ    PRETRACKER_FASTER_CODE  ; this extra code doesn't seem to be justified
         tst.w   d2
         bne.s   .wave_new_loop_pos_fits ; perfect fit for last loop
-
         ; partial fit only
-        subq.w  #1,pcd_inst_subloop_wait_w(a5)  ; why, oh why?
-
+        subq.w  #1,d5                   ; why, oh why?
+        ENDC
 .wave_new_loop_pos_fits
         move.w  d1,pcd_inst_loop_offset_w(a5)
 
 .wave_subloop_wait
+        move.w  d5,pcd_inst_subloop_wait_w(a5)
         move.w  d1,pcd_out_lof_w(a5)
 
         moveq.l #0,d1                       
         bra.s   .wave_load_sample_offset
 
 .wave_has_no_subloop
-        moveq.l #0,d1
-        move.b  pcd_wave_offset_b(a5),d1
-        beq.s   .loop_handling_done
-        tst.b   wi_allow_9xx_b(a3)
-        beq.s   .loop_handling_done
+        tst.w   d1
+        beq.s   .wave_loop_handling_done
 
         ; apply offset from pattern for sample without subloop
         lsl.w   #7,d1
@@ -2049,8 +2041,7 @@ pre_PlayerTick:
         add.l   pv_wave_sample_table(a4,d2.w),d1
         move.l  d1,pcd_out_ptr_l(a5)
 
-        ;move.w  #$FFFF,pcd_out_lof_w(a5)   ; this is already set to -1, no need to set it again
-.loop_handling_done
+.wave_loop_handling_done
 
 ; ----------------------------------------
 ; pitch handling